VaxSight : An Explainable Deep Learning Framework for Predicting the Severity of Post‐Vaccination Adverse Event Reports

While vaccines remain one of humanity's greatest shields against infectious disease, the health risks that can follow immunization cannot be ignored. Behind every adverse event report lies a patient story, making early risk identification crucial for safer, more personalized vaccination practices. This paper presents VaxSight, an interpretable deep learning framework designed to predict the severity of post‐vaccination adverse event reports using real‐world data from the Vaccine Adverse Event Reporting System (VAERS), comprising over 340,000 reports spanning 2020–2025. The framework integrates a small set of demographic and vaccine‐related attributes (age, sex, vaccine type, manufacturer, and dose series) with unstructured symptom text processed via TF‐IDF and truncated SVD. All fields that record or depend on the clinical outcome are excluded from the input features to prevent label leakage, and models are validated using both a held‐out test set and fivefold stratified cross‐validation with bootstrap 95% confidence intervals. Six models are evaluated: Random Forest, XGBoost, LightGBM, CatBoost, TabNet, and FT‐Transformer, alongside two proposed hybrid architectures. Among all models evaluated under identical conditions, CatBoost achieves the highest macro‐averaged F1‐score (0.6462) with an accuracy of 90.26%, followed by the FT‐Transformer (macro‐F1 0.6305, accuracy 90.48%). Cross‐validation confirms the stability of these results. To ensure transparency, SHAP analysis is incorporated, revealing that patient age and vaccine‐related features are the most influential predictors of report severity. Age‐stratified analysis shows that individuals aged 60 and above account for disproportionately higher medium (13.3%) and high (8.0%) severity rates compared to younger cohorts. By turning thousands of unstructured reports into prioritized, explainable severity alerts, VaxSight offers a scalable path toward faster and more transparent pharmacovigilance triage.

Authors

Institutions

Publication Details

Journal
Applied AI Letters
Published
2026-11-18
DOI
https://doi.org/10.1002/ail2.70059
Primary Topic
Pharmacovigilance and Adverse Drug Reactions
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

VaxSight : An Explainable Deep Learning Framework for Predicting the Severity of Post‐Vaccination Adverse Event Reports

Saha Reno, Tahsin Shuborna, Azizul Hakim Shuvo, Namira Tasnim
Applied AI Letters
Pharmacovigilance and Adverse Drug Reactions
article

VaxSight : An Explainable Deep Learning Framework for Predicting the Severity of Post‐Vaccination Adverse Event Reports

Saha Reno, Tahsin Shuborna, Azizul Hakim Shuvo, Namira Tasnim
article en

Abstract

While vaccines remain one of humanity's greatest shields against infectious disease, the health risks that can follow immunization cannot be ignored. Behind every adverse event report lies a patient story, making early risk identification crucial for safer, more personalized vaccination practices. This paper presents VaxSight, an interpretable deep learning framework designed to predict the severity of post‐vaccination adverse event reports using real‐world data from the Vaccine Adverse Event Reporting System (VAERS), comprising over 340,000 reports spanning 2020–2025. The framework integrates a small set of demographic and vaccine‐related attributes (age, sex, vaccine type, manufacturer, and dose series) with unstructured symptom text processed via TF‐IDF and truncated SVD. All fields that record or depend on the clinical outcome are excluded from the input features to prevent label leakage, and models are validated using both a held‐out test set and fivefold stratified cross‐validation with bootstrap 95% confidence intervals. Six models are evaluated: Random Forest, XGBoost, LightGBM, CatBoost, TabNet, and FT‐Transformer, alongside two proposed hybrid architectures. Among all models evaluated under identical conditions, CatBoost achieves the highest macro‐averaged F1‐score (0.6462) with an accuracy of 90.26%, followed by the FT‐Transformer (macro‐F1 0.6305, accuracy 90.48%). Cross‐validation confirms the stability of these results. To ensure transparency, SHAP analysis is incorporated, revealing that patient age and vaccine‐related features are the most influential predictors of report severity. Age‐stratified analysis shows that individuals aged 60 and above account for disproportionately higher medium (13.3%) and high (8.0%) severity rates compared to younger cohorts. By turning thousands of unstructured reports into prioritized, explainable severity alerts, VaxSight offers a scalable path toward faster and more transparent pharmacovigilance triage.

Applied AI LettersVol. 8(1)
Ahsanullah University of Science and Technology (BD)
Good health and well-being
Pharmacovigilance and Adverse Drug Reactions
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon